COCHONUT: Recognizing Complex Chords from MIDI Guitar Sequences

Chord recognition from symbolic data is a complex task, due to its strong context dependency and the large number of possible combinations of the intervals which the chords are made of, specially when dealing with dissonances, such as 7, 9, 13 and suspended chords. None of the current approaches deal with such complexity. Most of them consider only simple chord patterns, in the best cases, including sevenths. In addition, when considering symbolic data captured from a MIDI guitar, we need to deal with non quantized and noisy data, which increases the difficulty of the task. The current symbolic approaches deal only with quantized data, with no automatic technique to reduce noise. This paper proposes a new approach to recognize chords, from symbolic MIDI guitar data, called COCHONUT (Complex Chords Nutting). The system uses contextual harmonic information to solve ambiguous cases, integrated with other techniques, such as decision theory, optimization, pattern matching and rule-based recognition. The results are encouraging and provide strong indications that the use of harmonic contextual information, integrated with other techniques, can actually improve the results currently found in literature.